读书笔记：Fuxi: a Fault-Tolerant Resource Management and Job Scheduling System at Internet Scale

最新推荐文章于 2025-09-30 14:49:33 发布

原创最新推荐文章于 2025-09-30 14:49:33 发布 · 290 阅读

0 ·

CC 4.0 BY-SA版权

读书笔记专栏收录该内容

9 篇文章

订阅专栏

本文探讨了阿里云伏羲平台作为资源管理和作业调度系统的特点，包括增量资源管理协议、用户透明故障恢复机制及有效的故障节点检测机制。对比YARN，伏羲将任务与容器分离，应用Master直接控制资源生命周期，支持容器复用，采用基于局部性树的调度策略。

此文是关于阿里云伏羲平台的论文，一些感兴趣的点：

Fuxi：a resouce management and job scheduling system. (我感觉是基于YARN做的，很像YARN)

1, An incremental resource management protocol

2, A user-transparent failure recovery

3, A effective (faulty-node) detection mechanism and a mlti-level blacklisting schema

Fuxi (FuxiMaster, AppMaster, Tubo) <>YARN(ResourceManager, AppMaster, NodeManager)

Fuxi 与 YARN区别：

1，Fuxi seperates the notion of task(the application process that performs the actual work) and container(the unit of resource grant). Once an application master receives an grant , it explicitly controls its life-cycle and may reuse the container to run multiple tasks.

2,Lcality tree based scheduling.